which trained model is the most accurate

1adrianb / face-alignment

:fire: 2D and 3D Face alignment library build using pytorch

https://www.adrianbulat.com

BSD 3-Clause "New" or "Revised" License

7.03k stars 1.34k forks source link

which trained model is the most accurate #267

Open Zju-George opened 3 years ago

Zju-George commented 3 years ago

Silly question here but I really want to get higher accuracy.

1adrianb commented 3 years ago

The weights of each group are identical and they simply reflect the internal changes made inside pytorch. Beyond this 2DFAN will produce 2D landmarks, 3DFAN - 2.5D landmarks and 3DFAN + the depth model will produce 3D landmarks. The projection of the 3D landmarks into 2D correspond with the 2.5D predictions.

Zju-George commented 3 years ago

@1adrianb Thanks for your reply! Yeah, I've just tested it out and I see that.

Also, one thing I've noticed is that the eye landmarks being not aligned perfectly when the eye is closed. Do you have any plan to fine-tune the model on images where people closing their eyes?

1adrianb commented 3 years ago

Sorry for delay, yes, but I don't have an exact timeline unfortunately.

Zju-George commented 3 years ago

Sorry for delay, yes, but I don't have an exact timeline unfortunately.

@1adrianb If you could share your training code, I would be willing to try the fine-tune work.